Correctional Offender Management Profiling for Alternative Sanctions
Correctional Offender Management Profiling for Alternative Sanctions (COMPAS) is an algorithm that uses 137 characteristics to calculate a probability of whether criminals will relapse or not. The algorithm should be proprietary , so it should be unknown how it works exactly and COMPAS should be a black box .
criticism
A study (“Pro Publica” study) in 2016 showed, however, that COMPAS was wrong particularly frequently and in a certain direction when the skin color of those affected was included. People with "black skin color" were therefore incorrectly classified as at risk of relapse almost twice as often as people with "white skin color". Conversely, the risk of relapse from COMPAS was assessed as lower in people with “white skin”. When checked by uninvolved, not legally educated test persons, it was found that they achieve similarly good “ predictions ” about the recidivism of offenders as this program.
Similar software was developed to use an algorithm (Public Safety Assessment) to determine who has to deposit a cash deposit and who does not (see: US deposit system ).
Manufacturer
The program was developed by Northpointe. The company was renamed Equivant .
Loomis v. Wisconsin
In Loomis v. Wisconsin , a review of the ruling has been sent to the United States Supreme Court to overturn the Wisconsin State's previous judgment against Loomis . The submission criticized the use of the COMPAS software in the sentencing of Eric Loomis to six years in prison because it had violated the accused's right to due process.
However, the Supreme Court declined to handle the case on June 26, 2017.
Web links
- Christian Stöcker: People are as smart as the expensive machine , Spiegel-online from January 28, 2018.
- “Pro Publica” study - Machine Bias .
- Are programs better than people at predicting reoffending? . The Economist , Jan. 17, 2018.
Individual evidence
- ↑ Algorithms decide on imprisonment or parole , website of the Association of Administrative Judges from January 30, 2018.
- ↑ Christian Stöcker: People are as smart as the expensive machine , Spiegel-online from January 28, 2018.
- ↑ Study by Julia Dressel, Hany Farid: The accuracy, fairness, and limits of predicting recidivism (English)
- ↑ Loomis v. Wisconsin , Petition for certiorari denied, June 26, 2017.
- ↑ Mitch Smith: In Wisconsin, a Backlash Against Using Data to Foretell Defendants' Futures , June 22, 2016.
- ↑ State v. Loomis , Harvard Law Review , March 10, 2017.
- ^ STATE of Wisconsin, Plaintiff - Respondent, v. Eric L. LOOMIS, Defendant Appellant. , Supreme Court of Wisconsin, July 13, 2016.
- ^ Rebecca Wexler: When a Computer Program Keeps You in Jail .
- ↑ ORDER LIST: 582 US , June 26, 2017.
- ↑ Eric L. Loomis, Petitioner v. Wisconsin , October 12, 2016.
